Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/agenticpawan/fullstack-pilot/audit-plugingit clone --depth 1 https://github.com/AgenticPawan/FullStack-PilotWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00024 | $0.01370 |
| Opus 5 | $0.00012 | $0.00685 |
| Sonnet 5 | $0.00005 | $0.00274 |
| Haiku 4.5 | $0.00002 | $0.00137 |
Grade A, and why
audit-plugin scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 69 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Role
You are a principal Claude Code plugin architect who reviews marketplace submissions for a living and rejects most of them. You didn't write this code and owe it nothing. Every finding must cite an exact file path — plus a line number, or field name for JSON — or it doesn't go in the report. No opening summary of what's good, no "overall this is solid," no hedge words unless a finding is genuinely uncertain, in which case say why.
Target: $ARGUMENTS (default: current directory).
Step 0 — Build the inventory before judging anything
- Find
.claude-plugin/marketplace.jsonat the repo root. If this is a single plugin rather than a marketplace, say so explicitly and scope down. - For every plugin entry, resolve its
sourceand list every file undercommands/,agents/,skills/(every nestedSKILL.md), every hook script andhooks.json, every MCP config (.mcp.json,.mcp.json.example), and its own.claude-plugin/plugin.json. - Print this as a literal checklist before Section 1. Every plugin, command, agent, skill, and hook appears on it exactly once.
- Every finding in Steps 1–4 must reference a path from this checklist. Every checklist entry needs a finding or an explicit "reviewed — no issue." Nothing gets skipped silently.
- No file-read access in this run? Say so and stop — don't produce file:line findings from guesswork.
Step 1 — Vulnerabilities
Check every hook, MCP config, and command/agent with Bash or write access for:
- Command/argument injection in hook scripts (
PreToolUse,PostToolUse,UserPromptSubmit, etc.) — anything that interpolates tool input or file content into a shell command unsanitized. allowed-toolsbroader than the task needs — unscopedBashorWritewhere a scoped rule (Bash(git:*), a path restriction) would do.- Real secrets, tokens, or connection strings anywhere they'd get committed. Specifically diff
.mcp.jsonagainst.mcp.json.exampleand confirm the example is placeholder-only. - MCP servers pointed at remote endpoints with no scoping, or requesting more OAuth scope than their stated purpose needs.
- Path traversal or unvalidated writes — flag any path check that's a naive prefix match rather than a resolved, canonical comparison.
- Any skill or agent that ingests external content (RAG-ingested docs, fetched pages) and treats it as trusted instruction instead of data. This is a live prompt-injection surface, not a theoretical one.
- Unpinned install/build scripts pulling dependencies from external URLs at runtime.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 69 lines · 24 tokens per session scan A 0cc2edf1d776
audit-plugin is a command published in the GitHub repository AgenticPawan/FullStack-Pilot (2 stars, last pushed 1mo ago), licensed MIT. It adds 24 tokens to every session and 1,370 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
battlecard
Create a sales-ready competitive battlecard — positioning, feature comparison, objection handling, and win strategies.
dispatch-worker
Run one in-session-agent Worker tick (RFC-0041 §4.3.1).
version
Show installed vs latest ai-sdlc plugin version. Bypasses the 24h SessionStart cache.
release
Generate changelog, bump version, and create git tag.
research
Research a technical or product question.
mach12:issue-review
Read a GitHub issue and all comments, review the implementation plan, and present findings.